fix(extract): collapse byte-identical duplicate edges at emission (#3251)

graphify/extractors/engine.py

@@ -3,6 +3,7 @@
 
 import hashlib
 import importlib
+import json
 from graphify.extractors.base import _LANGUAGE_BUILTIN_GLOBALS, _file_stem, _make_id, _read_text
 from graphify.ids import normalize_id
 from graphify.extractors.models import LanguageConfig
@@ -5979,9 +5980,23 @@ def _scan_js_module_dispatch(n) -> None:
     # ── Clean edges ───────────────────────────────────────────────────────────
     valid_ids = seen_ids
     clean_edges = []
+    # Byte-identical duplicates collapse to one edge (#3251): a signature that
+    # annotates the same type twice (``def f(a: Path, b: Path)``) is ONE
+    # reference relationship at one location, but the per-occurrence emission
+    # loops above append it once per annotation — in every language block, since
+    # neither add_edge nor the raw appends de-duplicate. The copies carry zero
+    # information (build's dedup drops them anyway) and their only observable
+    # effect is tripping diagnose_extraction's exact_duplicate_edges health
+    # warning. Only edges whose ENTIRE payload is identical collapse; any
+    # differing field (source_location, context, metadata, …) keeps both.
+    _seen_edge_payloads: set[str] = set()
     for edge in edges:
         src, tgt = edge["source"], edge["target"]
         if src in valid_ids and (tgt in valid_ids or edge["relation"] in ("imports", "imports_from", "re_exports")):
+            payload = json.dumps(edge, sort_keys=True, default=str)
+            if payload in _seen_edge_payloads:
+                continue
+            _seen_edge_payloads.add(payload)
             clean_edges.append(edge)
 
     # Ruby mixins were collected during the node walk (before raw_calls existed);

tests/test_csharp_call_site_generic_args.py

@@ -213,7 +213,16 @@ def test_call_site_generic_args_appear_in_issue_repro(tmp_path):
         ),
     })
     izeta_refs = [tgt for src, tgt, _ctx in all_di_refs if src == ".Di()" and tgt == "IZeta"]
-    assert len(izeta_refs) >= 2, (
-        "s.AddScoped<IZeta, Box<IZeta>>() must link BOTH the outer IZeta "
-        f"and the inner IZeta (inside the Box<...> argument); got {izeta_refs!r}"
+    # The outer IZeta and the inner IZeta (inside Box<...>) are the same
+    # reference relationship at the same location, so since #3251 the two
+    # walk occurrences collapse into ONE edge at extraction (they were always
+    # collapsed by build, and the raw duplicate tripped
+    # diagnose_extraction's exact_duplicate_edges warning). The call-site bug
+    # this test guards emitted NO edge at all, so existence keeps the teeth.
+    assert len(izeta_refs) == 1, (
+        "s.AddScoped<IZeta, Box<IZeta>>() must link IZeta exactly once "
+        f"(one relationship, no duplicate edge); got {izeta_refs!r}"
+    )
+    assert (".Di()", "Box") in {(s, t) for s, t, _c in all_di_refs}, (
+        "the Box<...> argument itself must still link from the Di method"
     )

tests/test_duplicate_annotation_edges.py

@@ -0,0 +1,93 @@
+"""Exact-duplicate edge emission from repeated annotations (#3251).
+
+A signature that annotates the same type twice (``def f(a: Path, b: Path)``)
+used to emit one identical ``references`` edge per occurrence — same source,
+target, relation, source_location and context. build() drops the copy, but
+diagnose_extraction first counts it under ``exact_duplicate_edges`` and the
+run ends with an unexplained GRAPH HEALTH WARNING. Duplicates now collapse at
+extraction; anything differing in any field still survives.
+"""
+
+import collections
+
+from graphify.diagnostics import diagnose_extraction
+from graphify.extract import extract
+
+
+def _edge_counts(result):
+    return collections.Counter(
+        (e["source"], e["target"], e["relation"],
+         e.get("source_location"), e.get("context"))
+        for e in result["edges"]
+    )
+
+
+def test_python_repeated_parameter_annotation_emits_one_edge(tmp_path, monkeypatch):
+    """The issue's exact repro: two same-typed params, one references edge."""
+    monkeypatch.chdir(tmp_path)
+    (tmp_path / "sample.py").write_text(
+        "from pathlib import Path\n\n\n"
+        "def two_params(a: Path, b: Path) -> None:\n"
+        "    print(a, b)\n\n\n"
+        "def one_param(a: Path) -> None:\n"
+        "    print(a)\n",
+        encoding="utf-8",
+    )
+    r = extract([tmp_path / "sample.py"], cache_root=tmp_path)
+    counts = _edge_counts(r)
+    dupes = {k: v for k, v in counts.items() if v > 1}
+    assert dupes == {}, f"exact duplicate edges emitted: {dupes}"
+    assert counts[("sample_two_params", "path", "references",
+                   "L4", "parameter_type")] == 1
+    assert counts[("sample_one_param", "path", "references",
+                   "L8", "parameter_type")] == 1
+
+
+def test_csharp_repeated_parameter_annotation_emits_one_edge(tmp_path, monkeypatch):
+    """The same disease exists in every language block — C# as the witness."""
+    monkeypatch.chdir(tmp_path)
+    (tmp_path / "Svc.cs").write_text(
+        "public class Svc {\n"
+        "    public void Copy(Widget a, Widget b) { }\n"
+        "}\n"
+        "public class Widget { }\n",
+        encoding="utf-8",
+    )
+    r = extract([tmp_path / "Svc.cs"], cache_root=tmp_path)
+    dupes = {k: v for k, v in _edge_counts(r).items() if v > 1}
+    assert dupes == {}, f"exact duplicate edges emitted: {dupes}"
+
+
+def test_diagnose_reports_zero_exact_duplicates(tmp_path, monkeypatch):
+    """The health warning the issue hit is gone at the source."""
+    monkeypatch.chdir(tmp_path)
+    (tmp_path / "sample.py").write_text(
+        "from pathlib import Path\n\n\n"
+        "def two_params(a: Path, b: Path) -> None:\n"
+        "    print(a, b)\n",
+        encoding="utf-8",
+    )
+    r = extract([tmp_path / "sample.py"], cache_root=tmp_path)
+    summary = diagnose_extraction(r)
+    assert summary["exact_duplicate_edges"] == 0
+
+
+def test_differing_locations_and_contexts_survive(tmp_path, monkeypatch):
+    """Only byte-identical edges collapse: the same type referenced at two
+    locations, or under two contexts at one location, keeps every edge."""
+    monkeypatch.chdir(tmp_path)
+    (tmp_path / "sample.py").write_text(
+        "from pathlib import Path\n\n\n"
+        "def alpha(a: Path) -> None:\n"
+        "    print(a)\n\n\n"
+        "def beta(b: Path) -> Path:\n"
+        "    return b\n",
+        encoding="utf-8",
+    )
+    r = extract([tmp_path / "sample.py"], cache_root=tmp_path)
+    counts = _edge_counts(r)
+    # Two locations → two edges.
+    assert counts[("sample_alpha", "path", "references", "L4", "parameter_type")] == 1
+    assert counts[("sample_beta", "path", "references", "L8", "parameter_type")] == 1
+    # Same location, different context (parameter vs return) → both edges.
+    assert counts[("sample_beta", "path", "references", "L8", "return_type")] == 1